Agri-SAGE Generates Context-Aware Agricultural Advisories with LLMs

Agricultural advisory systems traditionally face a dilemma: static agronomic guidelines offer consistency but lack adaptability to in-season variability and dynamic uncertainties. Conversely, recent LLM-powered systems risk generating recommendations that are agronomically plausible but physiologically unconvincing. To address these limitations, researchers developed Agri-SAGE, a closed-loop framework. Agri-SAGE integrates retrieval-grounded multi-agent LLM reasoning with APSIM-based biophysical simulation. This combination allows the system to generate and rigorously validate agronomic advisories that are highly context-aware. The framework was evaluated using three reasoning approaches—Plan-and-Solve, Tree of Thoughts, and Reflexion—over a 10-year retrospective analysis. All three approaches significantly surpassed static Package-of-Practice baselines, with Tree of Thoughts achieving peak yields. Reflexion, while comparable in agronomic outcomes, proved more computationally efficient by leveraging cross-seasonal episodic memory.